Identification and observation of mechatronic systems including multidimensional nonlinear dynamic functions

Abstract

In this article a new approach on identification and observation of mechatronic systems is presented. This approach deals with complex nonlinear systems. These systems consist of a well-known dynamic subsystem and an unknown nonlinear dynamic subsystem with multiple inputs and a single output (MISO). The structure and the parameters of the dynamic subsystem are exactly known, whereas only rough knowledge about the structure and no knowledge about the parameters of the nonlinear dynamic subsystem is available. The identification algorithm is based on the Volterra theory. Orthonormal base functions are introduced for the purpose of parameter reduction. Furthermore the identification algorithm is enhanced with a normalized radial basis function network in order to approximate highly complex nonlinear static functions within the unknown subsystem. Finally the identification algorithm is implemented into an observer. In this manner the identification becomes practicable for mechatronic systems, e.g. nonlinear motion systems, where the output of nonlinear dynamic functions cannot be measured directly. The adaptation law of the identification algorithm has to be extended by an error transfer function in order to enable the learning process. The presented approach is illustrated by a simulation example.